[time-nuts] Frequency divider design critique request

Bruce Griffiths bruce.griffiths at xtra.co.nz
Wed Jul 16 05:56:48 EDT 2008

Hal Murray wrote:
> [Context is filtering on inputs from switches.]
> Could somebody tell me why we are doing anything fancy at all?  What's wrong 
> with just a simple pullup?
Pull down required because of thumbwheel switch encoding.
> I can think of two cases that might be interesting.
> One is signal integrity.  There might be enough crosstalk to cause troubles.  
> This is a 4 layer board.  Right?  Unless the traces from the switch to the 
> mux chip wrap around some high speed signal there isn't likely to be a 
> problem.
The 10MHz clock tracks have been routed well clear of everything else.
> I'm not a signal integrity wizard, but the ballpark is that you only need a 
> few trace widths of separation between agressor/source and victim/receiver.  
> This case is slightly ugly since the pullup is not low impedance like a 
> typical driver.
> So the question becomes how small a pullup do we need to maintain good signal 
> integrity?  Or how far from a nearby trace do we have to be with a given 
> pullup?
> If the coupling is primarily capacitive, then we have a C-R high pass filter. 
>  I'm not sure that's valid, but it is easy to analyze.  I should be smart 
> enough to work this out, but it's late.  HC is pretty slow.  AC might be fast 
> enough to make things interesting.  But this only matters if the switch 
> traces run parallel to a trace that is active for a significant length.
> What is the (ballpark) output impedance of a CMOS driver?  What's the input 
> capacitance on a CMOS gate?  (It's in parallel with the R, making a C-C 
> divider at high frequencies.)
10-20 ohms.
> The other possible complication is trying to keep a clean output signal when 
> the switch changes state.  I'm assuming we don't really care what happens 
> while the contacts are bouncing between switch settings.  If we do, then we 
> need serious switch debouncing and synchronization.  I don't think anybody 
> has mentioned this tangle.
RC filter followed by a Schmitt trigger IC (74HC14) can debounce the 
switch contacts effectively if RC product is large enough.
A shift register (clocked at 1MHz in this case)is required to 
synchronise each mux control input if synchronisation is required.
Using a shift register is slightly simpler than using a dual FF as only 
need to connect clock plus input and output.
> All the switch inputs I've worked with recently have been a simple 
> pullup/down.  They usually go into a small CPU using the internal pullups and 
> software debouncing.
No cpu available.
> I have heard war stories of cross talk on reset signals.  They often wander 
> all over the board and since they are slow, they often get overlooked when 
> checking for crosstalk.
Using an emitter follower to buffer the reset circuit RC node has led to 
VHF oscillation problems with capacitive load on emitter (track C) with 
collector and base at RF ground.


More information about the time-nuts mailing list